Stress assignment to polysyllabic words is the only source of inconsistency in reading aloud Italian, an orthography that is fully transparent. Italian has two main stress patterns for polysyllabic words—that is, penultimate stress, in which the penultimate syllable bears stress (e.g., par.TI.ta,Footnote 1 “match”), and antepenultimate stress, in which the antepenultimate syllable bears stress (e.g., BI.bi.ta, “drink”). Most of the time the position of stress is not predictable by rule.Footnote 2 To correctly assign stress to a word, an Italian reader must know the word. However, readers may assign stress on the basis of some statistical knowledge regarding the distribution of stress patterns in the language. In fact, the two main stress patterns have an asymmetrical distribution, with the majority of three- and more syllable words (around 80 %) bearing penultimate stress and far fewer words (around 18 %) bearing antepenultimate stress (Thornton, Iacobini & Burani, 1997).Footnote 3

Some studies on adult reading have indicated that stress assignment to polysyllabic words may be affected by the most frequent stress pattern in the language (stress dominance). Colombo (1992) reported an effect of stress dominance—that is, penultimate (dominant) stress words were read faster than antepenultimate (nondominant) stress words, but only when they were of low frequency (see also, for English, Rastle & Coltheart, 2000; for similar results on Italian adults with acquired language impairments, Colombo, Fonti & Cappa, 2004; Laganaro, Vacheresse & Frauenfelder, 2002).

However, polysyllabic word reading is also affected by stress neighborhood—that is, the proportion of words in the language that share the same stress pattern and the same final orthographic–phonemic sequence (e.g., the final sequence -ola is associated with the least frequent stress—on the first syllable—since it occurs predominantly in three-syllabic words such as PEN.to.la “pot” that bear antepenultimate stress). Burani and Arduino (2004) challenged the stress dominance effect on low-frequency word reading by showing that when dominant and nondominant stressed low-frequency Italian words were matched for the number of stress friends and enemies, they did not differ significantly for naming times and accuracy. Furthermore, both dominant- and nondominant-stress low-frequency words were easier to read when they had many stress friends than when they had many stress enemies. It was concluded that stress neighborhood is the most powerful determinant of stress placement on low-frequency words and that stress neighborhood information is a much more reliable cue of stress assignment than is stress dominance.

Not only in Italian, but also in English, orthographic sequences carry information for stress pattern, and are thus especially important for English adult readers. Words with reliable orthographic cues to stress assignment were processed more successfully than words with misleading cues in reading and lexical decision, indicating that readers can learn orthographic correlates to lexical stress that are located in the final syllable (Arciuli & Cupples, 2006; Kelly, Morris & Verrekia, 1998).

Stress assignment has been less investigated in developing readers, with the main interest being on testing how developing readers assign stress to nonwords. Recently, it has been shown that English-speaking children assigned the most frequent stress pattern (on the first syllable) to disyllabic nonwords, but progressively with age they attended more to orthographic cues (beginnings and especially endings) as predictors for stress position (Arciuli, Monaghan & Ševa, 2010). Similar results were obtained in a nonword-reading study with Italian children (Sulpizio, Boureux, Burani, Deguchi & Colombo, 2012).

The issue of how young readers assign stress to words has—to our knowledge—received very little attention. In a developmental study in Greek—a language with transparent orthography and lexical stress (similarly to Italian)—various sources of information were found to contribute to stress assignment. Studying word (and nonword) reading, Protopapas and Gerakaki (2009) reported that, when assigning stress, young readers were affected by three types of sources—that is, lexical information (word knowledge), visual–orthographic information (by processing of the written diacritic, which is placed over the vowel of the stressed syllable), and stress dominance (by which stress is assigned to the penultimate syllable; see also Protopapas, 2006).

In Italian, only two studies have investigated stress assignment to words, and they reported contrasting results. Paizi, Zoccolotti and Burani (2011) ran two experiments to test whether the reading accuracy of young readers with and without developmental dyslexia was affected by stress dominance and stress neighborhood. In the first experiment, Paizi et al. showed that, in reading low-frequency words, sixth-grade children with dyslexia were less accurate in reading antepenultimate (nondominant) stress than penultimate (dominant) stress words. Stress dominance, however, did not affect typically developing sixth graders. In the second experiment, the authors found that both typically developing children and children with dyslexia were equally affected by stress neighborhood, showing greater accuracy for low-frequency words with many stress friends than for low-frequency words with many stress enemies, irrespective of stress dominance. Recently, Sulpizio and Colombo (2013) reported that young Italian readers (second and fourth graders) were influenced by stress dominance when reading low-frequency words. However, stress dominance differently affected younger and older children, with its effect reaching significance with the younger, second-grade group only.

Thus, we currently have a puzzling picture: On the one hand, Paizi et al. (2011) tested children with dyslexia and typically developing sixth graders and showed that whereas stress neighborhood affects both groups in reading low-frequency words, only children with dyslexia are affected by stress dominance. On the other hand, Sulpizio and Colombo (2013) reported that younger readers are affected by stress dominance when reading low-frequency words. The two studies seem to indicate that the reliance on stress dominance changes with age and reading ability: Whereas for beginning readers and readers with scarce reading abilities dominance seems to be the stronger source for assigning stress to words that are not well known, more mature readers seem to increasingly rely on stress neighborhood—that is, on the similarity of stress between words that share orthographic endings.

However, the two existent studies on stress assignment by Italian developing readers not only investigated the performance of children of different ages and reading abilities, but also differed in the stimuli that were used: Whereas Paizi et al.(2011) used nondominant-stress words with consistent neighborhoods, Sulpizio and Colombo (2013) used nondominant-stress words with inconsistent neighborhoods, favoring the children’s tendency to make stress errors. Thus, currently we do not know whether younger typically developing readers than the ones studied by Paizi et al. may behave like older skilled readers, by mainly exploiting consistent stress neighborhoods with nondominant-stress words also, when they are given the opportunity.

The aim of the present study was to settle the contrasting findings by further investigating the influence of stress dominance and stress neighborhood on fourth graders and comparing their performance with that of adult readers by using crucial experimental contrasts. The following two reading-aloud experiments were designed to investigate the effects of stress dominance in interaction with both word frequency (Exp. 1) and stress neighborhood (Exp. 2) in both adult and fourth-grade readers. Experiment 1 crucially differed from the one run on fourth graders by Sulpizio and Colombo (2013) because, like Experiment 1 of Paizi et al. (2011) on older readers, stress neighborhood was kept large and consistent in both dominant- and nondominant-stress words. In Experiment 2, the effect of a consistent stress neighborhood on the reading of low-frequency words with contrasting stress patterns was investigated for the first time on fourth graders and compared to adults’ performance. In these conditions, if readers did not simply take advantage of a coarse statistical knowledge of stress dominance, but if instead more subtle knowledge of the statistical association of a word’s ending to a certain stress pattern has already been acquired as early as fourth grade, we would expect that stress neighborhood, but not stress dominance, might affect the word reading of both adults and children. Unlike all of the preceding studies with Italian young readers, in the present experiments we also investigated whether reading latencies, in addition to errors, are affected by stress properties.

Experiment 1: Stress dominance by word frequency

In Experiment 1, we examined the interaction between stress dominance and word frequency. Previous studies in Italian have shown that both young and adult readers are affected by stress dominance, but only when reading low-frequency words (Colombo, 1992; Colombo et al., 2004; Laganaro et al., 2002; Sulpizio & Colombo, 2013). In the present experiment, we aimed to examine whether or not the effect of stress dominance on low-frequency words is present when dominant- and nondominant-stress words have similar proportions of stress friends and enemies. Therefore, as opposed to the latter studies that investigated the stress dominance by word frequency interaction, but similarly to Paizi et al. (2011), dominant and non-dominant stress words were matched for the proportion of stress friends. It was expected that in such a condition, in which low-frequency words (with either dominant or non-dominant stress pattern) are given similar probabilities of being stressed correctly on the basis of their neighborhood, by having a larger proportion of stress friends (i.e., words with which they share ending and stress pattern) than enemies (i.e., words with the same ending but different stress pattern), the effect of stress dominance may disappear in both adults and fourth graders.

Experiment 1a: Children

Method

Participants

A group of 26 fourth graders (nine boys and 17 girls), 8.6–10.0 years old (mean age = 9.6 years, SD = 0.4) participated in the study. The children attended one elementary school in Rome. All of the children came from a middle socio-educational background and showed no evidence of sensory or neurological impairment. Only children with a scores above the 25th percentile on Raven’s Coloured Progressive Matrices (CPM; Pruneti et al., 1996) were included in the sample. Parents were informed about the screening activities and authorized their children’s participation in the study. The children had reading abilities within normal range, which was evaluated by means of a standardized for Italian reading achievement test (MT Reading test; Cornoldi & Colpo, 1998). Children were asked to read a text aloud within a 4-min time limit, and their reading speed (time in seconds per syllable) and accuracy (number of errors, adjusted for the amount of text read) were measured. The criterion for inclusion in the experimental group was a performance on the MT reading test within normal limits (defined as within 1 SD above or below the mean of the normative data from Cornoldi & Colpo, 1998) for reading speed and accuracy. On the MT reading test (Cornoldi & Colpo, 1998), the selected children made few errors on average (4.6; SD = 2.37, corresponding to about 2.07 % with respect to the total number of words in the passage); this corresponds to a mean z score of 0.08 (SD = 0.46) with regard to the normative values of children at the end of fourth grade. As for reading speed, children read the text at a mean rate of 0.30 s/syllable (SD = 0.61), which corresponds to a mean z score of 0.32 (SD = 0.13), indicating normal performance.

Materials

A list of 34 high-frequency (HF) and 34 low-frequency (LF) (Marconi, Ott, Pesenti, Ratti & Tavella, 1993) three- and four-syllable nouns was used. Half of the words in each frequency set (HF-LF) carried dominant stress, and half carried nondominant stress. All four word sets (see Appendix A) were matched for orthographic neighborhood size (N-size), length, mean bigram frequency, orthographic complexity, and initial phoneme (voicing and manner of articulation). Crucially, the words with both dominant and nondominant stress had more stress friends than enemies, as calculated on a frequency count of written Italian (Bertinetto et al., 2005).

Procedure

The stimuli were presented in three blocks, each composed of similar numbers of HF and LF words and words with dominant and nondominant stress. The presentation order of the blocks and of trials within each block was automatically randomized. A practice block contained eight words, half with dominant and half with nondominant stress.

The participants read aloud the stimuli that appeared in the center of the computer screen, with an interstimulus interval of 1,500 ms. Reaction times (RTs) in milliseconds were measured at the onset of pronunciation. Each stimulus disappeared at pronunciation or after 3,500 ms. A voice key connected to the computer measured RTs in milliseconds at the onset of pronunciation. RTs were collected using E-Prime software (Psychology Software Tools, Pittsburgh, PA; www.pstnet.com). The experimenter noted the naming errors or apparatus failures on the fly. Due to the ease of identifying word stress for each native Italian speaker, the stress errors, too, were noted by the experimenter on the fly.

Results

Invalid trials due to technical failures or responses that exceeded the time limit (3.6 % of the data points) were not analyzed. The results are presented in Figs. 1a (mean RTs of items named correctly) and 1b (percentages of pronunciation and stress errors). Pronunciation errors occurred when the participant did not accurately pronounce the word at the segmental level (i.e., because of phoneme substitutions, omissions, insertions or transpositions, hesitations, stuttering, or false starts), whereas stress errors consisted only in the incorrect placement of stress (by either assigning dominant stress to nondominant-stressed words or nondominant stress to dominant-stressed words). Separate analyses were conducted on pronunciation errors and stress errors

Fig. 1
figure 1

Experiment 1a, Stress Dominance × Word Frequency for children. (a) Mean reaction times (RTs) by participants. (b) Percentages of pronunciation and stress errors. Error bars represent standard errors.

Full size image

The RTs of correct responses and errors were analyzed using mixed-effects models. The models were fitted using the lmer function (Baayen, Davidson & Bates, 2008) in R software (version 3.0.1). Frequency (high vs. low) and Stress Dominance (dominant vs. nondominant) were used as fixed factors, whereas Participants and Items were treated as random factors.

In the naming time analysis, the RTs were log-transformed to reduce the skewness of the data, and p values were calculated using the MCMC procedure, sampling 10,000 times (Baayen et al., 2008). The model showed a main effect of word frequency (t = 3.10, beta = 0.092, SE = 0.02, pMCM < .001, 95 % CI = [0.04, 0.14]) but no effect of stress dominance (t = −0.34, beta = −0.010, SE = −0.01, pMCM = .7, 95 % CI = [−0.06, 0.47]). No Stress × Frequency interaction (t = 0.63, beta = 0.027, SE = 0.02, pMCM = .5, 95 % CI = [−0.05, 0.10]) was found.

The numbers of errors that children made were similar to those reported in previous studies with Italian children (cf., e.g., Paizi et al., 2011) and were large enough to run statistical analysis. The mixed-effects logistic regression analyses on pronunciation errors paralleled those on RTs: Participants were more accurate when reading high- rather than low-frequency words (z = 3.10, beta = 1.394, SE = 0.449, p = .001, 95 % CI = [0.51, 2.27]); no other effect reached significance (stress dominance, z = −0.57, beta = −0.300, SE = 0.525, p = .5, 95 % CI = [−1.51, 0.44]; Word Frequency × Stress Dominance, z = 0.53, beta = 0.349, SE = 0.652, p = .5, 95 % CI = [−0.74, 1.8]).

The analysis on stress errors did not show any significant effect (word frequency, z = 0.006, beta = 16.30, SE = 2809.51, p = .9, 95 % CI = [−5,488, 5,505]; stress dominance, z = 0.005, beta = 15.05, SE = 2809.5, p = .9, 95 % CI = [−5,489, 5,520]; Word Frequency × Stress Dominance, z = −0.005, beta = −13.58, SE = 2809.51, p = .9, 95 % CI = [−5,491, 5,518]).

Experiment 1b: Adults

Method

Participants

A group of 22 (16 females and six males) university students, native Italian speakers (mean age = 24.1 years, SD = 2.5) participated in the experiment.

Materials and procedure

These were the same as in Experiment 1a.

Results

Invalid trials (3.6 % of data points) were discarded from the analyses. Figure 2 reports the mean RTs of items named correctly in the different experimental conditions.

Fig. 2
figure 2

Experiment 1b, Stress Dominance × Word Frequency for adults: Mean reaction times (RTs) by participants. Error bars represent standard errors.

Full size image

The model on log RTs revealed a main effect of word frequency (t = 2.08, beta = 0.111, SE = 0.053, pMCM = .03, 95 % CI = [0.01, 0.21]), with participants being faster when reading high- rather than low-frequency words. Neither the effect of stress dominance (t = −0.66, beta = −0.035, SE = 0.053, pMCM = .5, 95 % CI = [−0.13, 0.06]) nor the Word Frequency × Stress Dominance interaction reached significance (t = −0.16, beta = −0.011, SE = 0.075, pMCM = .8, 95 % CI = [−0.15, 0.13]). Due to ceiling performance in accuracy (overall: 0.6 % errors), variance in errors was too low to be analyzed.

Experiment 2: Stress dominance by stress neighborhood

The results of Experiment 1 showed that both children’s and adults’ reading was affected by word frequency, but not by stress dominance, when polysyllabic words of similar frequency but different stress were matched for stress neighborhood—that is, for the proportions of stress friends versus enemies. In Experiment 2, we investigated the effect of stress neighborhood in interaction with stress dominance on low-frequency word reading. A word’s ending (the vocalic nucleus of the penultimate syllable plus the last syllable) is correlated with a certain proportion of words that carry either dominant or nondominant stress (Burani & Arduino, 2004; Colombo, 1992). For instance, a large proportion (about 81 %) of Italian words ending in -oro carry dominant stress, whereas a smaller proportion (about 19 %) of words with the same ending are stressed on the antepenultimate syllable. Yet most words (about 77 %) ending in -ola carry nondominant stress, and a much smaller proportion carry dominant stress (about 23 %). Thus, the word BAMbola (doll) carries nondominant stress, but has many stress friends (i.e., words sharing the same final sequence and stress pattern). In contrast, the word pistOLA (gun), although carrying dominant stress, has many stress enemies (i.e., words with the same final sequence but a different stress pattern).

Italian adults have been shown to be affected by the composition of stress neighborhood in reading low-frequency words (Burani & Arduino, 2004). However, whether stress neighborhood also affects the accuracy and reading times of children as young as fourth graders has not yet been investigated. If fourth graders are sensitive to the existence of stress neighbors—as is suggested by the effect of stress neighborhood on older children’s accurate stress assignment (Paizi et al., 2011)—that is, if children rely on the most informative units (word endings) for stress placement—then an overall influence of stress neighborhood, with faster and more accurate reading of words with many stress friends than for words with many stress enemies, would be expected. In addition, if stress neighborhood is the major determinant of stress assignment to unfamiliar stimuli for children of this age (see also Arciuli et al., 2010; Sulpizio et al., 2012) and adults (Burani & Arduino, 2004), then stress neighborhood should overrule stress dominance and affect both nondominant and dominant stress words (Burani & Arduino, 2004), in both fourth graders and adults.

Experiment 2a: Children

Method

Participants

A group of 28 (12 females and 16 males) fourth graders, 8.7–10.2 years old (mean age = 9.3 years, SD = 0.4), with reading abilities within normal range, participated in the experiment. None of them had participated in Experiment 1a. All of the children attended one elementary school in Rome, came from a middle socio-educational background, and showed no evidence of sensory or neurological impairment. The criteria for inclusion in the experimental group were the same as were described for Experiment 1. On the MT reading test (Cornoldi & Colpo, 1998), the selected children made few errors on average (4.54, SD = 2.42, corresponding to about 2.04 % with respect to the total number of words in the passage); this corresponds to a mean z score of 0.09 (SD = 0.47) with regard to the normative values of children at the end of fourth grade. As for reading speed, children read the text at a mean rate of 0.31 s/syllable (SD = 0.65), which corresponds to a mean z score of 0.3 (SD = 0.14), indicating normal performance

Materials

Four sets of 14 three- and four-syllable low-frequency words were presented, varying in stress dominance (dominant, nondominant) and stress neighborhood (many friends, many enemies). The words in each of the two sets (i.e., dominant–many friends vs. nondominant–many enemies, nondominant–many friends vs. dominant–many enemies) had the same endings, selected so as to maximize the difference in the proportions of friends versus enemies (Table 1). The sets were matched on word frequency (Bertinetto et al., 2005), N-size, length, orthographic complexity, mean bigram frequency, and initial phoneme (see Appendix B).

Table 1 Percentages of stress friends in types and tokens for words with dominant and nondominant stress used in Experiment 2
Full size table

We added to the list 56 familiar (medium- to high-frequency) filler words, with a variety of final sequences different from the experimental ones, so as to avoid encouraging any strategic processing due to the repetition of final sequences. Half of the filler words were stressed on the penultimate, and half on the antepenultimate syllable, and all were matched with the experimental items on length.

Procedure

The 112 items were presented in four blocks of 28 stimuli each, half experimental and half filler words, half with dominant and half with nondominant stress. The orders of blocks and of the within-block trials were randomized. The practice block contained ten items (five with dominant and five with nondominant stress). The general procedure was as in Experiment 1.

Results

Invalid trials (4.1 % of data points) were discarded from the analyses. The results are presented in Figs. 3a (mean RTs of items named correctly) and 3b (percentages of pronunciation and stress errors).

Fig. 3
figure 3

Experiment 2a, Stress Dominance × Stress Neighborhood for children. (a) Mean reaction times (RTs) by participants. (b) Percentages of pronunciation and stress errors. Error bars represent standard errors.

Full size image

Mixed-effects models were conducted on the log-transformed RTs of correct responses, pronunciation errors, and stress errors. Stress dominance (dominant vs. nondominant) and stress neighborhood (many friends vs. many enemies) were used as fixed factors, whereas Participants and Items were treated as random factors.

The model on log RTs showed a main effect of stress neighborhood (t = 2.81, beta = 0.115, SE = 0.04, pMCM = .002, 95 % CI = [0.04, 0.18]): Participants were faster when reading words with many friends than when reading words with many enemies. However, no effect of stress dominance (t = 0.16, beta = 0.006, SE = 0.04, pMCM = .8, 95 % CI = [−0.06, 0.84]), nor a Stress × Frequency interaction (t = −0.72, beta = −0.042, SE = 0.05, pMCM = .4, 95 % CI = [−0.14, 0.06]), was found.

The mixed-effects logistic analysis on stress errors paralleled that on RTs: The model showed a main effect of stress neighborhood (z = 2.45, beta = 1.506, SE = 0.61, p = .014, 95 % CI = [0.31, 2.69]), but no effect of stress dominance (z = 1.34, beta = 0.85, SE = 0.63, p = .17, 95 % CI = [−0.38, 2.08]) and no Stress Neighborhood × Stress Dominance interaction (z = −0.88, beta = −0.72, SE = 0.81, p = .3, 95 % CI = [−2.3, 0.86]).

The mixed-effects logistic analysis on pronunciation errors did not show any significant effect (stress neighborhood, z = 1.50, beta = 0.57, SE = 0.38 p = .13, 95 % CI = [−0.17, 1.31]; stress dominance, z = −0.26, beta = −0.10, SE = 0.40, p = .8, 95 % CI = [−0.88, 0.68]; Stress Neighborhood × Stress Dominance, z = −0.38, beta = −0.21, SE = 0.54, p = .7, 95 % CI = [−1.26, 0.84]).

Experiment 2b: Adults

Method

Participants

A group of 24 native Italian speakers (ten females and 14 males), collaborators at ISTC-CNR, participated in the experiment. Their mean age was 28.2 years old (SD 4.0). None of them had participated in Experiment 1b.

Materials and procedure

These were the same as in Experiment 2a. Two experimental blocks of 56 trials each were presented.

Results

Invalid trials (2.5 % of the data points) were discarded from the analyses. Figure 4 reports mean RTs of items named correctly in the different conditions.

Fig. 4
figure 4

Experiment 2b, Stress Dominance × Stress Neighborhood for adults: Mean reaction times (RTs) by participants. Error bars represent standard errors.

Full size image

The model on log RTs revealed a main effect of stress neighborhood (t = 1.90, beta = 0.035, SE = 0.018, pMCM = .04, 95 % CI = [0.003, 0.067]), No further effect reached significance (stress dominance, t = −0.12, beta = −0.005, SE = 0.042, pMCM = .8, 95 % CI = [−0.08, 0.74]; Stress Neighborhood × Stress Dominance, t = 0.02, beta = 0.0005, SE = 0.026, pMCM = .9, 95 % CI = [−0.04, 0.05]). Due to ceiling performance in accuracy (overall: 2.1 % errors), the variance was too low to allow for an error analysis.

Discussion

The results of the present study show that stress neighborhood—that is, the proportion of words that share the same stress pattern and the same final orthographic–phonemic sequence—is a strong determinant of stress assignment. Importantly, information concerning stress neighborhood is already available to fourth-grade readers, and when stress friends in the neighborhood are many, this seems to outweigh the role of stress dominance.

In Experiment 1, we examined whether both the naming times and accuracy of young Italian readers are affected by stress dominance in interaction with word frequency. This interaction was assessed by means of stimuli that had been matched for stress neighborhood. For fourth-grade readers, as well as for adults, when stimuli had similarly high proportions of stress friends, neither facilitation for dominant-stress words nor an interaction between stress dominance and word frequency was apparent. That is to say, when stimuli were matched for stress neighborhood (the proportion of stress friends), the interaction between stress dominance and frequency was ruled out, with no more facilitation for low-frequency words with dominant stress (e.g., Colombo, 1992; Sulpizio & Colombo, 2013).

Experiment 2 demonstrated that typically developing Italian children in the fourth grade rely efficiently on word endings to process polysyllabic stimuli: Low-frequency words whose neighborhood was constituted by a prevalence of stress friends were read faster and more accurately than words with a prevalence of stress enemies, irrespective of their stress pattern (dominant or nondominant), as in adult readers.

The results of the present experiments are consistent with recent findings reported for nonword reading in both English and Italian typically developing young readers (Arciuli et al., 2010; Sulpizio et al., 2012), and for word reading with Italian typically developing children and children with dyslexia (Paizi et al., 2011), and they contribute to detailing the development of sensitivity to statistical properties of the language in Italian readers.

Taken together, the results of our experiments allow us to shed new light on the process of stress assignment in reading Italian aloud. On the basis of the present results, it appears that early in development Italian readers assign stress by relying on subtle statistical information, such as stress neighborhood, at least when they have to read low-frequency words. This finding adds to the increasing effects of stress neighborhood on nonword reading that have been reported by both Sulpizio and Colombo (2013) and Sulpizio et al. (2012), with fourth graders already behaving similarly to adults in the capacity of picking up information coming from stress neighborhood in order to assign stress to unknown stimuli.

Our data show that stress assignment is affected by lexical knowledge (as demonstrated by the word frequency effect) and stress neighborhood. However, we do not argue that Italian readers are fully impermeable to stress dominance. Paizi et al. (2011) reported that sixth graders with dyslexia were less accurate when reading nondominant- (antepenultimate) stress than when reading dominant- (penultimate) stress words. Moreover, Sulpizio and Colombo (2013) reported that young children, when reading low-frequency words with nondominant stress and many stress enemies, tended to assign the wrong stress (i.e., dominant stress) to these words, and that this tendency was not any more significant with increasing age (see also Sulpizio et al., 2012; for similar results in English, see Arciuli et al., 2010). It is possible that in some cases, young Italian readers may assign stress by relying on stress dominance. This is, however, restricted to cases in which readers have limited lexical sources of information available, such as readers with developmental dyslexia and very young children (e.g., second graders) (cf. also Colombo et al., 2004, and Laganaro et al., 2002, who studied adults with acquired disturbances of reading).

Overall, Italian skilled readers seem to rely on stress neighborhood to assign stress to low-frequency polysyllabic words. A possible account for the present pattern of results might be offered by the connectionist framework for stress assignment in English proposed by Arciuli and colleagues (Arciuli et al., 2010; Ševa, Monaghan & Arciuli, 2009). Arciuli et al.’s (2010) developmental model, trained incrementally using age-appropriate reading materials, learned to map orthography onto stress position, successfully simulating young readers’ sensitivity to the statistical properties of their language. Specifically, Arciuli et al.’s (2010) model accounted for increasing reliance on orthographic cues for stress assignment in children’s reading aloud. According to this model, as readers grow older and are exposed to age-appropriate reading materials, they become progressively sensitive to statistical cues revealed by orthography, especially those contained in word endings.

A more detailed account of our results can be offered by the CDP++ model of reading, which is implemented for adults’ reading of English (Perry, Ziegler & Zorzi, 2010). The model implements a detailed stress system: The phonological output buffer includes two distinct mechanisms—that is, phonological output nodes and stress output nodes—with the latter nodes being responsible for stress assignment. The stress output nodes receive activation from both the lexical and sublexical routes and combine the two sources of information through competitive interaction. The sublexical route is a two-layer connectionist network that maps graphemes onto phonemes and the orthographic input onto a stress pattern. During training, the network may learn to associate specific orthographic cues—such as word endings—with a certain stress pattern. In the model, the stress neighborhood effect may emerge at the level of the phonological output buffer: In the case of words with many stress enemies (e.g., piSTOla, “gun”), the stress output nodes will receive contrasting information from the two routes (e.g., penultimate stress activated by the lexical route, and antepenultimate stress activated by the -ola sequence in the sublexical route). As a consequence, the stress system will take more time (and possibly make more errors) to activate the correct stress pattern, especially when lexical activation is weak, as in the case of low-frequency words. Finally, in accordance with the absence of any stress dominance effect in our results, the current architecture of the CDP++ predicts no advantage (at least in terms of reading times) for words with the dominant stress pattern in the language.

To conclude, the results of the present study show that word frequency (Exp. 1) and stress neighborhood (i.e., the proportions of stress friends and enemies) (Exp. 2) affect polysyllabic word reading in terms of naming times and accuracy for readers of different ages and reading experience, both adults and children in the fourth grade. In our experiments, we found no evidence for a stress dominance effect. It could be argued that stress friendship, other than lexical knowledge, is the main source of the information exploited by Italian readers to assign stress to polysyllables: Word endings as cues for stress position can be easily identified (and become reliable), when they recur in several words in the language (i.e., when the stress neighborhood of friends is sufficiently dense; see Sulpizio, Arduino, Paizi & Burani, 2013) and are a better stress predictor than stress dominance. When such cues to stress position are available, readers—both adults and, more importantly, children—attend to these cues rather than the dominant stress pattern. In short, readers are sensitive to the statistical properties of their language and rely efficiently on orthographic–phonological predictors for stress position.